cs.CVMar 27, 2026

Face2Parts: Exploring Coarse-to-Fine Inter-Regional Facial Dependencies for Generalized Deepfake Detection

Authors: Kutub Uddin, Nusrat Tasnim, Byung Tae Oh

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Multimedia data, particularly images and videos, is integral to various applications, including surveillance, visual interaction, biometrics, evidence gathering, and advertising. However, amateur or skilled counterfeiters can simulate them to create deepfakes, often for slanderous motives. To address this challenge, several forensic methods have been developed to ensure the authenticity of the content. The effectiveness of these methods depends on their focus, with challenges arising from the diverse nature of manipulations. In this article, we analyze existing forensic methods and observe that each method has unique strengths in detecting deepfake traces by focusing on specific facial regions, such as the frame, face, lips, eyes, or nose. Considering these insights, we propose a novel hybrid approach called Face2Parts based on hierarchical feature representation (HFRHFR) that takes advantage of coarse-to-fine information to improve deepfake detection. The proposed method involves extracting features from the frame, face, and key facial regions (i.e., lips, eyes, and nose) separately to explore the coarse-to-fine relationships. This approach enables us to capture inter-dependencies among facial regions using a channel-attention mechanism and deep triplet learning. We evaluated the proposed method on benchmark deepfake datasets in both intra-, inter-dataset, and inter-manipulation settings. The proposed method achieves an average AUC of 98.42% on FF++, 79.80% on CDF1, 85.34% on CDF2, 89.41% on DFD, 84.07% on DFDC, 95.62% on DTIM, 80.76% on PDD, and 100% on WLDR, respectively. The results demonstrate that our approach generalizes effectively and achieves promising performance to outperform the existing methods.

Figures & tables

Fig. 1: Key concepts of the proposed method: (a) The conventional deepfake detection, which relies solely on facial artifact analysis. (c) In contrast, the proposed method, which leverages hierarchical information across multiple levels, coarse (frame), medium (face), and fine-grained (lips, left eye, right eye, and nose), aims to improve deepfake detection performance.
Fig. 2: Categories of deepfake detection methods: (a) conventional methods, which rely on handcrafted features, (b) deep learning-based methods that automatically learn discriminative representations from data, and (c) hybrid approaches that combine traditional cues with deep learning models to improve detection performance.
Fig. 3: Examples of deepfake generation using different manipulation techniques: (a) Face swapping, where the source face is replaced with a target face in the FF++ [ 7 ] /FSW dataset, and (b) lip synchronization, where facial lip movements are manipulated using different audio tracks in the PDD [ 40 ] dataset.
Fig. 4: Performance analysis comparing feature representations: This study compares face-only and lips-only features with the proposed HFR approach. The results demonstrate that hierarchical feature representations provide superior effectiveness for deepfake detection across the FF++ [ 7 ] /FSW, FF++ [ 7 ] /NT, and PDD [ 40 ] datasets.
Fig. 5: Architecture of the proposed method: The proposed framework is composed of three sequential phases. In Phase 1, an encoder network is employed to extract hierarchical feature representations from multiple facial regions, including the full frame, face, lips, left eye, right eye, and nose. In Phase 2, a deep triplet learning strategy combined with a CA-MLP mechanism is introduced to model and exploit interrelationships among regional features, yielding discriminative embeddings organized into two clusters. Finally, in Phase 3, the learned embeddings are passed to a classifier for final prediction.
Fig. 6: Key facial regions extracted using Dlib: An 81-point facial landmark detector was employed to accurately localize facial regions and capture detailed facial information. The colored rectangles indicate the selected facial regions used for extracting coarse-to-fine-grained hierarchical features.
Fig. 7: Block diagram of the proposed CACA -MLP embedding network, which models relationships among multi-level hierarchical features extracted from the frame, face, left eye, right eye, lips, and nose regions.
Fig. 8: Illustrative examples of real and deepfake frames (from left to right) from the FF++ [ 7 ] , CDF2 [ 26 ] , DFD [ 39 ] , and PDD [ 40 ] datasets. Deepfake faces in these datasets were generated using different synthesis models.
DatasetsSubjectsFocus# of Samples (Real/Fake)ManipulationsTrain-Test
FF++ [ 7 ]977Random1000/5000FaceSwap, FaceShift, NeuralTextures3600, 700
CDF1 [ 26 ]N/ACelebrities408/795FaceSwap1027, 176
CDF2 [ 26 ]59Celebrities590/5639FaceSwap5632, 597
DFD [ 39 ]28Random363/3068Not disclosed, probably FaceSwap2744, 687
DFDC [ 39 ]66Random1131/4119Not disclosed, probably FaceSwap3675, 1575
DTIM [ 41 ]64Random320/320FaceSwap448, 192
TABLE I: Detailed overview of the dataset: Including dataset name, number of subjects, target group characteristics, sample count, and manipulation types.
Fig. 9: Deepfake detection performance across different information levels: (a) frame-level AUC and (b) video-level AUC evaluated on eight benchmark datasets.
Fig. 10: Hierarchical feature representations are extracted using a pretrained CLIP model across three levels: level-1 (frame), level-2 (face), and level-3 (lips, left eye, right eye, and nose regions).
Fig. 11: t-SNE visualization of hierarchical feature embeddings after triplet learning. The top row shows embeddings using only level-2 features, which are commonly adopted in deepfake detection, while the bottom row illustrates embeddings obtained using multi-level hierarchical features from level-1 to level-3.
Fig. 12: Performance comparison of different classification methods using Face and HFR features. (a)-(c) depict frame-level AUC, while (d)-(f) illustrate video-level AUC for FF++ [ 7 ] /DF, CDF1 [ 26 ] , and PDD [ 40 ] , respectively.
Fig. 13: Examples of failure cases in the inter-setting evaluation. (a) CDF2 [ 26 ] real, (b) CDF2 [ 26 ] fake, (c) DFD [ 39 ] real, and (d) DFD [ 39 ] fake samples. Detection results are reported using ViT-L/14, as it achieved superior performance compared to the other evaluated detectors.
Figure 15
Figure 16
Figure 17
More in this topic

Deepfake Detection

Deepfake detection identifies manipulated or AI-generated media to support trustworthy communication, identity verification, and forensic analysis.

See all 241 papers

Explore similar work

CardsList
Oct 2, 2026

Interpretable Deepfake Detection in Videos via Explicit Forensic Features and Temporal Modeling

Deepfake detection in videos remains challenging, as manipulated content may appear visually consistent at the frame level while exhibiting subtle temporal inconsistencies. This paper introduces an interpretable deepfake detection framework that models spatially and temporally coherent facial features in video sequences. Unlike end-to-end deep models relying on implicit representations, the proposed approach explicitly encodes physically grounded forensic cues, enabling transparent analysis and improved multi-dataset generalization. The pipeline transforms videos into identity-consistent facial trajectories, segments them into fixed-length temporal windows, and represents each frame using 68 structured descriptors spanning four complementary domains: photometric, textural, geometric, and compression-based features. These descriptors provide a compact multi-domain representation of manipulation artifacts and are processed by a Long Short-Term Memory (LSTM) network to capture temporal dependencies and subtle irregularities. Evaluation on four benchmark datasets, FaceForensics++, Celeb-DF v2, a curated subset of the DeepFake Detection Challenge (DFDC), and DeeperForensics, yields strong and consistent F1-scores of 98.0%, 91.0%, 97.6%, and 96.2%, respectively. The approach also demonstrated a good cross-dataset generalization, providing a robust and interpretable solution for video deepfake detection.

Chahira Benhama, Mohand Saïd Allili, Assia Hamadene

LARIVIA laboratory Dep. of Computer science and Eng. University of Quebec in Outaouais Gatineau, QC, Canada · LARIVIA laboratory and Mixed Research Unit INRS-UQO University of Quebec in Outaouais Gatineau, QC, Canada

Video ForensicsDeepfake DetectionExplainable Deepfake DetectionDigital Forensics
AskPDF
Apr 29, 2026

Whether, Which, and Whose: Solving the Triple Challenge of Deepfake Proactive Forensics in Multi-Face Scenarios

Unlike single-face forgeries, deepfakes in complex multi-person interaction scenarios (such as group photos and multi-person meetings) more closely reflect real-world threats. Although existing proactive forensics solutions demonstrate good performance, they heavily rely on a "single-face" setting, making it difficult to effectively deal with the problems of deepfake detection, localization, and source tracing in complex multi-person environments. In this paper, we propose a Deep Attributable Watermarking Framework (DAWF) tailored to multi-face proactive forensics, which establishes an isolated identity attribution space. This spatial isolation ensures that multiple independent tracing signals can coexist within a single image and be successfully anchored to their respective identity instances. Crucially, we propose a selective regional supervision loss to suppress cross-face interference, guiding the decoder to focus exclusively on the manipulated facial regions. DAWF unifies image-level detection, instance-level localization, and identity-level source tracing, successfully achieving the "whether, which, and whose" forensic goals of determining whether an image is manipulated, which face is forged, and whose identity is tampered with. Extensive experiments on challenging multi-face datasets demonstrate robust triple-forensic performance, achieving an AUC of 0.91, an F1-score of 0.87, and a BER of 0.54%. The code is available at https://github.com/vpsg-research/DAWF.

Lei Zhang, Zhiqing Guo, Dan Ma +2

School of Computer Science and Technology, Xinjiang University, Urumqi, 830017, China · Xinjiang Multimodal Intelligent Processing and Information Security Engineering Technology Research Center, Urumqi, 830017, China · College of Computer Science and Electronic Engineering, Hunan University, Changsha, 410082, China

Digital ForensicsImage Forgery LocalizationDeepfake DetectionRobust Watermarking
AskPDF
Mar 2, 2026

Deepfake Forensics Adapter: A Dual-Stream Network for Generalizable Deepfake Detection

The rapid advancement of deepfake generation techniques poses significant threats to public safety and causes societal harm through the creation of highly realistic synthetic facial media. While existing detection methods demonstrate limitations in generalizing to emerging forgery patterns, this paper presents Deepfake Forensics Adapter (DFA), a novel dual-stream framework that synergizes vision-language foundation models with targeted forensics analysis. Our approach integrates a pre-trained CLIP model with three core components to achieve specialized deepfake detection by leveraging the powerful general capabilities of CLIP without changing CLIP parameters: 1) A Global Feature Adapter is used to identify global inconsistencies in image content that may indicate forgery, 2) A Local Anomaly Stream enhances the model's ability to perceive local facial forgery cues by explicitly leveraging facial structure priors, and 3) An Interactive Fusion Classifier promotes deep interaction and fusion between global and local features using a transformer encoder. Extensive evaluations of frame-level and video-level benchmarks demonstrate the superior generalization capabilities of DFA, particularly achieving state-of-the-art performance in the challenging DFDC dataset with frame-level AUC/EER of 0.816/0.256 and video-level AUC/EER of 0.836/0.251, representing a 4.8% video AUC improvement over previous methods. Our framework not only demonstrates state-of-the-art performance, but also points out a feasible and effective direction for developing a robust deepfake detection system with enhanced generalization capabilities against the evolving deepfake threats. Our code is available at https://github.com/Liao330/DFA.git

Jianfeng Liao, Yichen Wei, Raymond Chan Ching Bon +3

Shenzhen Technology University, Guangdong, China · Singapore Institute of Technology, Singapore, Singapore · The University of Hong Kong, Hong Kong, China +1

Deepfake DetectionDigital Image ForensicsVLM Adaptation
AskPDF